http://www.genetic-programming.com/johnkoza.htm
Scientific Research Interests—John R. Koza
Our main research interest is automatic programming (also called
program synthesis or program induction)—that is, getting computers to
solve problems without explicitly programming them.
This goal can be accomplished using the technique of genetic
programming (of which I am considered the inventor). Genetic
programming is an automated method for creating a working computer
program from a high-level problem statement of a problem. Genetic
programming performs automatic program synthesis using Darwinian
natural selection and biologically inspired operations such as
recombination, mutation, inversion, gene duplication, and gene
deletion. Old Chinese saying says "animated gif is worth one
megaword," so click here for short tutorial of "What is GP?" For
information about the rapidly growing field of genetic programming,
visit www.genetic-programming.org and www.genetic-programming.com
While proof of principle ("toy") problems are occasionally useful for
tutorial or introductory purposes, we believe that it is time for
fields of artificial intelligence and machine learning to start
delivering non-trivial results that satisfy the test of being
competitive with human performance. Accordingly, our criterion for
undertaking new research is that, if the anticipated outcome of the
research effort is achieved, it can be argued (on some reasonable
basis) that the result created by genetic programming is competitive
with human-produced results. Competitiveness with human performance
can be established in a variety of ways. For example, genetic
programming may produce a result that is slightly better, equal, or
slightly worse than that produced by a succession of human researchers
working on an well-defined problem over a period of years. Or, genetic
programming may produce a result that is equivalent to an invention
that was patented in the past or that is patentable today as a new
invention. Or, genetic programming may produce a result that is
publishable in its own right (i.e., independent of the fact that the
result was mechanically generated). Or, genetic programming may
produce a result that wins or ranks highly in a judged competition
involving human contestants. There are examples using genetic
programming in all four categories and we have been produced at least
one example in three of the four categories. Fourteen are described in
detail in the Genetic Programming III: Darwinian Invention and Problem
Solving book and Human-Competitive Machine Intelligence videotape For
additional discussion, see human-competitive machine intelligence.
Specifically, our recent research work involving genetic programming
currently emphasizes
automated synthesis of analog electrical circuits,
automated synthesis of controllers,
automated synthesis (reverse engineering) of metabolic pathways
(networks of chemical reactions),
automated synthesis of antennas,
automated synthesis of genetic networks,
problems in computational molecular biology,
various other problems involving cellular automata, multi-agent
systems, mathematical algorithms, and other areas of design, and
using genetic programming as an automated "invention machine" (for
creating new and useful patentable new inventions).
There are now a number of instances where genetic programming has
automatically produced a computer program that is competitive with
human performance. (See our criteria for human-competitive results and
a list of human-competitive results by clicking on human-competitive
machine intelligence). The fact that genetic programming can evolve
entities that are competitive with human-produced results suggests
that genetic programming may possibly be used as an "invention
machine" to create new and useful patentable inventions. In this
connection, evolutionary methods, such as genetic programming, have
the advantage of not being encumbered by preconceptions that limit
human problem-solving to well-traveled paths.
In late July 1999, Genetic Programming Inc. started operating a new
1,000-node Beowulf-style parallel cluster computer consisting of 1,000
Pentium II 350 MHz processors and a host computer. Genetic Programming
Inc. has also operated (starting in early 1999) a 70-node Beowulf-
style parallel cluster computer consisting of 533 MHz DEC Alpha
microprocessors and a host computer. The new 1,000-Pentium system is
called the Tera-COTS computer (since it has capacity of about a
teraflops and is a beowulf-style customer computer made of "commodity
off-the-shelf" [COTS] parts). Click here for technical discussion of
parallel genetic programming and building the 1,000-Pentium Beowulf-
style parallel cluster computer.
All of the above-mentioned 21 human-competitive results were obtained
using computers that were substantially smaller than the new 1000-
Pentium computer mentioned above. Fifteen of these 21 human-
competitive results were obtained on a 1995-vintage parallel computer
system composed of 64 PowerPC 80 MHz processors with a spec95fp rating
that is 1/60 of that of the new 1000-Pentium machine. Five of these
results were obtained on a 70-Alpha machine (whose spec95fp rating is
1/9 of that of the 1000-Pentium machine). One of these human
competitive results were obtained with a 1994-vintage machine (whose
spec95fp rating is 1/1,320 of that of the 1000-Pentium machine).
Because of its increased computational power of the new 1000-Pentium
machine, we expect that it will produce additional human-competitive
results.
Genetic programming has 16 important attributes that one would
reasonably expect of a system for automatic programming (also called
program synthesis or program induction). Genetic programming has seven
important differences from other approaches to machine learning and
artificial intelligence.
--
You received this message because you are subscribed to the Google Groups
"Algorithm Geeks" group.
To post to this group, send email to algogeeks@googlegroups.com.
To unsubscribe from this group, send email to
algogeeks+unsubscr...@googlegroups.com.
For more options, visit this group at
http://groups.google.com/group/algogeeks?hl=en.
